Third Party Interventions and the Duration of Intrastate Conflicts

Patrick M. Regan Department of Political Science

Binghamton University Email: pregan@binghamton.edu

607-777-2167 An earlier version of this paper was presented at the workshop on the Economics of Political Violence, Princeton University, May, 2000 and the Annual meetings of the American Political Science Association, Aug 28-Sept. 1, 2000. This project was partially supported by the Post Conflict Fund of the World Bank. Thanks to Tom Brunell and the anonymous reviewers.

2

Abstract Recent research has begun to focus on the role of outside interventions in the duration of civil conflicts. Assuming that interventions are a form of conflict management, we would hold ex ante expectations that they would reduce a conflict’s expected duration. In this paper I test hypotheses relating the type and timing of outside interventions to the duration of civil conflicts. The data incorporate 150 conflicts during the period 1945-1999, 101 of which had outside interventions. In those 101 cases there were a total of 1036 individual interventions. Using a hazard analysis the results suggest that third party interventions tend to extend expected durations rather than shorten them. Specifically, interventions that beget opposing interventions increase the expected duration dramatically over the baseline expectations, as do nearly any military or economic intervention. The only aspect of the strategy for intervening that reduces the likelihood that a conflict will end in the next month is that it be biased in favor of either the opposition or the government. In effect, neutral interventions are less effective than biased ones.

3

The management of internal conflicts has become increasingly important since the end of

the Cold War. The UN alone has seen a multifold increase in the number of peacekeeping

operations (Bobrow and Boyer, 1997; Khanna et al, 1998), while various other national and

multinational efforts have taken on increasing importance in efforts to influence civil

conflicts (Regan, 2000). Many questions remain about the effectiveness of efforts to

manage or ameliorate the consequences of internal conflicts. This study contributes to our

understanding of the effect of third parties on the duration of civil conflicts. With few

exceptions most of the empirical analyses have focused on the role of diplomatic efforts to

resolve internal disputes. It is only recently that scholarship has begun to focus on the role

of outside military and/or economic interventions in the expected duration of civil conflicts.

Presumably when outside parties intervene in an internal conflict at the core of their

motivation is some form of conflict management. Interveners may prefer their ally to

prevail, but one would think that prevailing at an acceptable cost and in a reasonable time

frame would be critical to an effective outcome.

In this paper I will elaborate more fully on the role of third parties in the duration of

civil conflicts1. First I will review the literature on the termination of civil conflicts, and

more specifically the factors that influence their duration. I will follow this with a

theoretical argument outlining the process by which outside actors influence the length of

time a conflict can be expected to last. Empirical analyses on the population of conflicts

from 1944 to 1999 with and without interventions will then be used to evaluate hypotheses

derived from my theoretical arguments. Finally I discuss the results of my analysis and

make suggestions for further research and future policy. Very briefly, the results suggest

4

that unilateral interventions tend to lengthen the expected duration of a conflict, but that

biased interventions are associated with shorter expected durations relative to neutral ones.

In general, however, most interventions appear to prolong rather than reduce the expected

duration of a conflict.

Literature Review

Previous research into the factors that effect the expected duration of civil wars is rather

sparse, nor is there much on the broader topic of the duration of wars in general. Bennett

and Stam (1996) present one of the first systematic analyses on the expected duration of

interstate war. Their conclusion was that a mix of realpolitik explanations and strategic

decisions by the principles prosecuting the war had the greatest impact on a conflict’s

expected duration. For example, using a duration model they find that the balance of

capabilities and the war fighting strategy are strong predictors of the expected time until the

conflict ends. Other contributing factors include the level of democracy and the extent of

domestic political repression. Furthermore, they find that war itself is not duration

dependent, that is, the amount of time that it is ongoing does not influence its expected

future duration.

Regan and Stam (2000) test a duration model of interstate conflict, focusing on the

role of conflict management efforts as predictors of expected durations. They specify a

curvilinear model, which posits that the probability of a conflict ending is a function both of

efforts to manage the conflict and the timing of those efforts. Their evidence demonstrates

1 By civil conflicts I refer to armed internal conflicts that experience at least 200 fatalities. This threshold is considerably short of the 1000 fatalities per year required by the Correlates of War criteria, and reflects an

5

that diplomatic management efforts make important contributions to the resolution of

interstate disputes, but that diplomacy is not equally successful throughout the life cycle of

a conflict. That is, the relationship between third party mediation and a conflict’s duration

is non-linear with respect to the timing of the diplomatic initiatives. In effect they

demonstrate that the strategy for intervening incorporates more than simply the

characteristics or behaviors of the mediator. The concept of a strategy also involves the

timing of the intervention.

The literature on the duration of civil wars is even more recent and even less

comprehensive. Licklider (1995; 1993) has examined the termination of civil conflicts,

arguing in effect that the decision to settle is a function of internal capabilities, which can

be influenced by external interventions. His research design, however, does not permit a

direct interpretation of expected durations, given an ongoing conflict. Walter’s (2000)

analysis of civil wars in the post WWII period suggests that to maintain a stable peace once

a settlement is reached requires confidence from both sides that disarmament agreements

will be complied with and that their opponent will not reinitiate the conflict. At the end of a

fully implemented peace agreement that stops a civil war, the government generally has the

upper hand because of its ability to rearm. This results in reluctance on the part of the

opposition to agree to the terms of a negotiated agreement. We saw this recently in

Northern Ireland where the IRA resisted disarmament allegedly out of fear of a Loyalist

rearmament on terms unfavorable to the IRA. According to Walter, outside interventions

are necessary to maintain the stability of a negotiated settlement by imposing costs on the

side that rearms. Regan’s (1996; 2000) analysis demonstrates that under certain conditions

outside interventions into civil wars can facilitate the end of the violent aspects of the

interest in what Ted Gurr referred to as “organized conspiricies” and “internal warfare” (Gurr, 1970).

6

conflict. But none of the studies mentioned directly answers the critical question of

whether outside interventions shorten or prolong the duration of civil wars.

Regan examined whether conflicts with outside interventions tended to last longer

than those without interventions (2000). The evidence suggests that they do, but the

inferences that one is able to draw from this analysis do not give a sufficient picture of the

relationship between interventions and expected durations. What I demonstrated is that

longer running conflicts tend to have more outside interventions, but the research design

cannot discriminate between the cause and effect. One possible explanation for this

evidence could be that third parties choose to intervene in long running conflicts, rather

than the intervention extending the length of time until a settlement. And if one makes the

assumption that interventions are a form of conflict management, and therefore the first

preference of the intervener is to bring the conflict to an end, then this interpretation is

consistent with the evidence. A more sophisticated duration model, with data capable of

explicitly testing the role of outside interventions is necessary to identify the role of

interventions in a conflicts expected duration. Three recent studies of civil war durations

begin to give us a sense of the factors that lead to longer rather than shorter wars, and in

particular the role of outside actors in influencing the expected duration of a civil conflict

(Elbadawi & Sambanis, 2000; Lindsay and Enterline, 1999; Collier, Hoeffler, Soderbom,

1999).

Elbadawi and Sambanis develop a formal theoretical treatment of the conditions that

contribute to the expected duration of civil wars. Using a micro-economic maximization

model he shows that external support for the combatants can effect the costs of continuing

the war, that the growth in rebel forces should have a direct influence on its course, and that

7

ethnic fractionalization should make rebel mobilization easier and increase a wars expected

duration. If supported by empirical evidence, there are at least two important implications

from this work that can inform theoretical development and potentially influence the policy

process. First, if his arguments are correct, the increased duration that results from

fractionalization of a society along ethno-linguistic lines would suggest that ideological

conflicts could be expected to have shorter durations that ethnically-based conflicts. This

would be consistent with speculations by Kauffman (1996), that ethnically based conflicts

are more difficult to resolve than other forms of civil war. Second, his model suggests that

particular types – or strategies -- of external interventions would appear to reduce the length

of a conflict, and this points directly to specific policy initiatives. An inference from

Elbadawi and Sambanis’s model is that an external intervention that addresses issues of

poverty will make rebel mobilization more difficult, thereby shortening the length of the

war. The fractionalization of societies along ethno-linguistic lines is not a necessary

condition that results from socio-cultural diversity, but rather a function of the degree of

discrimination along these lines (Gurr, 1993).

Collier et al (1999) estimate a hazard model of the duration of civil wars, using of

sample of 45 civil wars taken from the Correlates of War database. They find support for

the notion that ethnic fractionalization has a strong influence on the expected duration of a

conflict. Interestingly, their results suggest that the relationship is non-linear, with

homogeneous and highly fractionalized societies having shorter civil wars than moderately

fractionalized societies. There are two potential explanations for the effects of high

homogeneity or heterogeneity: 1) it may influence the governments ability to divide

loyalties within the opposition, and 2) it is consistent with the idea that civil wars in

8

homogeneous or highly heterogeneous societies tend to organize along ideological lines.

Their results also suggest that civil wars are not duration dependent. Collier et al suggest

that the balance of capabilities between the government and the rebels should influence the

duration of a civil war, however their evidence does not support a such a relationship. They

fail, however, to take into account the effect of external actors on the relative balance of

capability. In effect, Elbadawi and Sambanis’s arguments about external interventions

extend the theoretical foundations by which we can think about the results of Collier et al

results; accordingly a more complete specification by Collier et al should include the effect

of interventions on the capabilities of the actors.

Using a sample of civil wars somewhat larger than Collier et al, though also taken

from the Correlates of War Civil War database, Lindsay and Enterline (1999) test a hazard

model of civil war duration. They articulate a model incorporating a number of contextual

variables, including but not limited to ethnic homogeneity and characteristics of any outside

intervention. Of their results that approach conventional levels of statistical significance,

third party support for the government is associated with an increased expected duration of

the conflict, and taken by itself, a balanced intervention increases the probability that the

conflict will end at any given point in time. Ethnic homogeneity, however, appears to be

unrelated to the expected duration of a conflict, though “identity wars” have longer

durations than ideological ones. Their evidence provides a contrasting view of the role of

the balance of capabilities on a conflict’s duration from that articulated by Collier and his

colleagues. By making the assumption that an intervention that supports the government

tends to create a greater imbalance of capabilities, the Lindsay and Enterline results suggest

that it is a balance of forces that leads to shorter wars and an imbalance to longer ones.

9

There are a number of common themes to these studies that point toward further

theoretical development and empirical analysis: demographics, the balance of capabilities

between the government and opposition, and the strategy for prosecuting a conflict are all

related to the duration of a conflict. The decision to continue fighting can be thought of in

terms of the costs and expectations associated with continued conflict, and this calculus is

influence by the ability to mobilize and the resulting balance of capabilities. In the

following section I will present a theoretical argument from which we can think about

strategic and contextual variables and their effect on the duration of civil wars, from which

I articulate testable hypotheses.

Intervention Strategies and the Duration of Intrastate Conflicts

My focus here is specifically on the role of outside interventions in the expected duration of

internal conflicts. I make a general assumption that outside interventions into internal

conflicts are a form of conflict management, and therefore attempt to control the hostilities

rather than exacerbate them. If the goal of an intervention can be primarily thought of in

terms of conflict management, then ceteris paribus, we would hold ex ante expectations that

interventions should reduce the duration of a conflict. This assumption is not without

controversy, even though it is useful for framing the role of interventions. For example, if

we assume that states intervene in an effort to reduce the duration, magnitude, or violence

associated with the conflict, then we would expect that on average they would be effective

at achieving these outcomes. The alternative is to view interventions as a form of inter-

nation influence designed to achieve a much broader range of outcomes, only one of which

10

might look like conflict management. Supporting the rebel movement with military aid, for

instance, might not appear to be an attempt at conflict management, while similar support

for the government might. That is, an intervention on behalf of the opposition is designed

to alter the pre-conflict status quo ante. In contrast, when an outside actor intervenes on

behalf of the government we can assume that this is an attempt to restore the pre-conflict

status quo ante. If a disparity in capabilities leads to an earlier settlement, then balancing

the capabilities in favor of the government could be expected to shorten the conflict's

duration and therefore act as a method of short-term conflict management (Balch-Lindsey

and Enterline, 2000; Elbadawi and Sambanis, 2001). Ultimately whether or not outside

interventions are a useful form of conflict management is an empirical question that can be

answered with the data.

To achieve their objectives an intervener must manipulate the costs of continued

fighting, the benefits from settlement, and do so in a way that convinces both sides that

settling now is preferable to the same outcome at some future time. Some might think of

this as using an intervention to create a 'ripe' moment (eg. Haass, 1990). In other words to

be successful both sides to the conflict must come to the conclusion that, at least at this

time, continued fighting presents too low an expected payoff. In decision-theoretic terms

an intervention is trying to maximize the expected utility of each of the actors for settling

now versus continued fighting until an expected victory. In effect, a conflict reflects a

bargaining situation where information is gained through successive moves. Outside

interventions effect the transmission of this information and its content. Most importantly,

an intervention will affect each actor’s estimate of their chances for victory by altering the

balance of capabilities required to sustain the fight.

11

The willingness to settle a conflict today versus continuing to fight until victory is a

function of expectations of future victory and current and anticipated costs. The balance of

capabilities affects the combatants’ estimation of the probability that victory can be

achieved, both at time ‘t’, and at some future period, ‘tn’. Likewise, costs, which are a

monotonically positive function of time, are to some degree related to the balance of

capabilities. Outside interventions can manipulate these expectations by initiating a biased

intervention (ie, on behalf of either the government or the opposition forces), which can

involve the user of force by the intervener.

Under these conditions the choice faced by the parties to the conflict involve trying

to maximize their expected payoff from their participation in the war, such that:

[ ( ) ( ) ( )( ) ]BCBCBCEU defeattn

defeat

tnvictorytn

victory

tnsettlet

n

t

settle

t

n

t

settle

tpp +−++−+=∑∑

=

10

where p is the probability of victory on the battlefield. Cst and Bs

t, respectively, are the costs

and benefits associated with a settlement in the current time period, which would be a

function of the cohesiveness of the rebel group (and/or the ruling coalition) and the

concessions that the opponent is willing to make to stop the fighting. Settling now – or at

any time between now and the time until victory – is judged relative to the anticipated cost

and benefits (Cvtn & Bv

tn) from being victorious and the probability that each side will win.

Costs always increase with increments of time, implying that Ct < Ctn. Furthermore, we can

assume that victory is preferred to a negotiated settlement, and therefore Bst < Bv

tn. This

implies that actors get more in future victory than they can through a current negotiated

settlement. The two critical conditions most effectively manipulated by outside

interventions are the respective expectations of victory and the costs associated with the

12

conflict. The strategy – or modality – of the intervention has an important impact on these

expectations.

The notion of a strategy incorporates the target of the intervention, its timing and

makeup, and whether or not to use force on behalf of the client. Intervening on behalf of

either the government or the opposition has the immediate effect of altering the balance of

capabilities. And since costs and estimates of victory will vary with time, the timing of the

intervention will contribute to the willingness of settle now versus continuing the fight.

Importantly, timing has a disproportionately large effect on the opposition because its

ability to mobilize will be a function of battlefield effectiveness. Relative strength on the

battlefield, moreover, affects reservation points and concession at the negotiating table. In

general without an intervention the government will be more powerful relative to the

opposition group, and therefore an intervention on behalf of the opposition should bolster

their expectations for victory, particularly if the intervention comes early in the conflict.

If we think of two sides calculating the probability of winning the conflict, and from

this estimate determining the value of a settlement now on terms less favorable than they

would obtain in victory, then the closer the capabilities come to parity the more likely that

1) they will have some reasonable expectation of victory, and 2) that there will be

significant error in this estimation. Outside military support for the government under these

conditions should in the norm shift the balance of capabilities toward preponderance, and

therefore the opposition might consider that the potential terms possible in negotiation have

a higher utility than the outcome of further military contest. In other words, an

intervention on behalf of the government should lead to shorter conflicts.

13

An intervention on behalf of the opposition should, in the norm, shift the balance of

capabilities toward parity. Under these conditions the rebel leadership would have an

increased expectation of their chances for victory. The closer the estimate gets to even odds

the more likely there is to be “forecasting errors” on the part of the rebels (Collier and

Hoeffler, 2000), and given a fixed level of grievance the intervention should encourage the

rebel movement to fight on and eschew negotiations, at least at the present time. Two

factors point toward this inference. First, as the balance of capabilities moves toward parity

the rebels can at minimum expect to extract greater concessions at the negotiating table.

The government, under conditions of parity, will have to make greater concession in order

to buy off the rebel movement because of the rebels increased expectations for gaining

concessions commensurate with their capabilities. If the government is unwilling or

incapable of making concessions -- as they might be in an ideological conflict – the rebels

will continue to fight. This might lead to what some have described as a hurting stalemate

(Zartman, 2000). Second, in light of outside support for their movement the rebels are

more likely to incorrectly estimate their chances for victory. Interventions in support of

rebel movements are more difficult than those in support of a government for reasons of

logistical efficiency and international norms, yet the net effect on a rebel movement is

greater than the same intervention supporting a government. In effect, a small intervention

supporting the opposition may increase their capabilities by a considerable amount, thereby

bolstering their expectations, even though not convincingly weighing the balance of

capabilities in their favor. This would lead the rebel leadership to press for greater demands

and be more resolute in their decision to fight.

14

Moreover, the timing of an intervention should significantly influence the duration

of a conflict through its effect on expectations. At the early stages of a conflict opposition

movements will generally be small, lightly armed, and lacking the crucial support of the

general public. Participation in opposition movements is costly and it often takes a small

committed group to get the movement rolling (DeNardo, 1983; see also Collier and

Hoeffler, 2000). So in the early stages we might anticipate that an armed opposition

movement is rather fragile and more susceptible to military defeat or early accommodation.

But after a period of consolidation by the opposition forces they will have higher – maybe

even inflated – estimates of their prospects for victory, and increased levels of resolve to

press on toward victory. Therefore, an intervention that takes place early in an armed

rebellion should have a considerably greater impact on the future course of the conflict. For

example, an intervention supporting the government early on will make the prospects for an

opposition victory appear more dire and lead them to either settle for a compromise

outcome or endure the overwhelming capabilities of the government. Alternatively, an

early intervention on behalf of the opposition will facilitate the organizational capabilities

of the opposition, contribute to the recruitment of additional troops, and increase estimates

that they can hold out for victory. In short, early interventions should have a significant

contribution to the expected length of a conflict, and the target of an early intervention will

determine whether it contributes to a shorter or longer conflict. This leads to the first two

hypotheses:

H1: Unilateral interventions that support the government early in the conflict will

shorten the expected duration of a conflict; and

15

H2: Unilateral interventions that support the opposition early in the conflict will

lengthen the expected duration of a conflict.

If the interest is in the effect of outside interventions on the time until the

termination of a conflict, then the strategy of the intervention is important. Strategy is a

multifaceted concept. It is used here to refer to the mix of military and economic

instruments, as well as the manner in which they are employed to achieve the outcome.

Either military or economic assistance can be given in large quantities or can be provided

incrementally, and these increments can be used to support the government or the

opposition. As mentioned above, the target of the intervention should have an independent

effect on the duration of a conflict, though the mix and amounts should also directly

influence the duration. The use of force by an intervener is a critical aspect of the strategy.

By actively fighting on behalf of a client an intervener directly alters the balance of

capabilities in favor of the side it supports. The opponent, therefore, would have to

calculate its expected payoff in terms of settling or fighting in light of having to engage

both its adversary and its adversary’s ally. It would also follow that the timing of the use of

force would influence the tempo of the participant’s calculations, with an early use of force

having a greater impact on expected durations than a later use.

Much of our theoretical expectations turn on the relative utilities each side has for

various outcomes. We can see this by focusing on expectations for victory. If we assume

that the decision to settle now versus continuing to fight is a function of their subjective

16

estimate of the probability of victory and their utility for various outcomes2, then if you can

always negotiate now, each side will choose to continue fighting when

p (Uwin) + (1-p) (Ulose) > Uneg

where p is the probability of victory, Uwin is the utility derived from victory, Ulose is the

utility from a loss, and Uneg is the utility from a negotiated settlement. Solving for p gives

the conditions under which the opposing sides will sue for peace.

losewin

loseneg

UUUU

p−−

>

In other words, the rebels will press on for victory when the ratio of what they stand to gain

from a negotiated settlement relative to victory is less than the probability of victory. Since

time is an important component of the costs of war (and therefore the utility of victory) the

timing of an outside intervention will have a discernible impact on the decision process.

Furthermore, opposing interventions – unless of dramatically unequal quantities –

would tend to maintain the status quo balance of relative capabilities, albeit at higher

absolute levels. Both sides would anticipate that they have increased prospects for victory

or a significant utility from holding out at the bargaining table for greater concessions from

the opponent. In general opposing interventions prolong a conflict’s expected duration.

This leads to three additional hypotheses:

H3: The use of force by an intervener will shorten the expected duration of a

conflict;

2 I change here from the use of costs and benefits to a more general use of utilities. One can think of utilities over outcomes as the sum of costs and benefits that reflect the value attributed to settlement, victory, or defeat.

17

H4: An early intervention that employs force will shorten a conflict relative to a late

use of force by an intervener;

H5: Interventions that attract counter-interventions will increase the expected

duration of a conflict

Strategies for intervening go beyond the notion of military efforts by states

supporting one side in a conflict. In other words the use of military instruments are not the

only mechanism available to third parties (Regan, 1996; 2000; Elbadawi, 1999; Elbadawi &

Sambanis, 2000). In fact if Elbadawi’s analysis is correct, military interventions may not

always be the best way to influence each side’s utility of settling now versus continuing on

until victory. We can see this most clearly by focusing on some of the causes of ethno-

political conflict, and the benefits that can accrue from settling.

For example, the extent of ethno-linguistic fractionalization is linked theoretically to

the onset and duration of civil wars (Elbadawi, 1999; Elbadawi & Sambanis, 2000), as well

as empirically (Collier et al, 1999). Military interventions would appear to have little direct

influence on the extent of societal fractionalization, beyond that of allowing rebel

movements to portray a higher likelihood of prevailing and therefore organize more

supporters into the movement. But ethno-linguistic fractionalization can be influenced by

outside interveners, just not by a reliance on military instruments. Fractionalization is not

an inherent aspect of a diverse society, but rather a function of the relative distribution of

resources (eg. Gurr, 1993). The more maldistributed the resources in an ethnically diverse

society the greater the level of fractionalization. And the more highly fractionalized the

18

society the greater the benefits from prevailing in a conflict. As a result mobilization is

made easier by the offer to redistribute resources upon victory (Lichbach, 1995). Therefore

one approach to shortening a civil war would be to intervene in a way that offers benefits to

the rebels from settling now versus holding out for victory at some future time. This will

increase the expected value of a negotiated settlement, reduce costs (which are a positive

function of time), and retard the ability to effectively mobilize forces (which would

decrease estimates of the probability of a rebel victory).

The strategy for intervening in a way that reduces the effects of ethno-linguistic

fractionalization must focus on the potential benefits from settling now versus fighting until

victorious, or the probability of victory. Multilateral interventions may serve this role by

being 1) neutral (see Regan, 2000; Diehl, 1993), and 2) distributing resources rather than

weapons. For example, the utility for victory declines as the degree of the maldistribution

of resources declines. If opposing forces can gain resources without incurring the costs of

conflict, their utility for settling now increases. Furthermore, as the benefits to be gained

from victory decline it should be harder for the opposition to recruit fighters, thereby

lowering the probability of victory. If we assume that neutral interventions are more likely

to distribute resources in a manner that increases equality, we should expect that:

H6: Neutral or multilateral interventions will lead to shorter expected durations

than unilateral interventions supporting either the government or the opposition.

19

Research Design

A useful way to think about the effect of interventions on a conflict’s duration is as an

intervention taking place at a discrete point in time. As a result of an intervention the

conflict either remains at the status quo condition or moves to an alternative state, which we

will call the termination of the conflict. The concern, therefore, is with the effect of these

discrete events on the timing of the transition between outcome states. The approach to

testing such effects is to use a duration, or hazard, model (Blossfeld and Rohwer, 1995;

Bennett, 1999; Box-Steffensmeier and Jones, 1997). Hazard models allow us to determine

the likelihood of a transition to state t`i, Given it is at state t0, for a series of explanatory

variables. Moreover, to fully capture the effect of an intervention we need to know the

probability of a transition from state t0 to t`i given an intervention at some point tn, where t0

< tn < ti.

There are different approaches for incorporating the effects of explanatory variables

into a duration model. One is to hold the covariates constant at a fixed point and examine

the impact of that event on expected durations. Doing so reduces the amount of useful

information associated with the causal process, and assumes that there is not variation in

effect on outcomes within a case over time. An alternative method is to employ a hazard

model where the explanatory variables vary as a function of time within a particular case.

This class of variables is called time varying covariates (TVC) (Bennett, 1999; Bennett and

Stam, 1996; Regan and Stam, 2000; Blossfeld and Rohwer, 1996).

The use of hazard models and TVCs is adopted largely from the medical sciences,

and posits a causal relationship much like that of a patient with a disease, a medical

intervention, and the expected time until the patient dies from the disease. The patient may

20

receive treatment at time tn (as well as subsequent treatments) and we can count the days

(months or years) until the patient passes on. Some patients may receive weekly

treatments, some monthly, some only once, and some never get treated. Furthermore, the

dosage administered may vary over time and across patients.

This analogy to civil conflicts and interventions is not unreasonable. The conflict is

conceptually similar to the patient, the intervention to the treatment, and the conflict either

remains active (alive) or terminates at some point ti (the patient dies). Because I have no

prior expectations about the duration dependence of civil conflicts, I adopt a Weibull

specification that allows this to be tested. 3 Without any covariates, the basic functional

form of the hazard rate λ (t) using a Weibull specification is:

)( 1)( t p

pt λλλ−

=

The function λ(t) is the hazard rate at some time t. The parameter p on the right hand side

of the model represents the duration dependence estimate. The parameter p determines the

shape of the hazard function, thereby accounting for duration dependence. When p=1, there

is no duration dependence, and the hazard rate λ(t) equals the constant rate l. When 0 < p <

1, the hazard rate decreases monotonically over time. When p > 1, the hazard rate increases

monotonically, though not necessarily linearly. Covariates (independent variables) can be

added into the model as influences on the hazard rate by specifying:

e iΧ−= βλ

3 There are alternative specifications of hazard models (eg. Cox proportional hazard), each of which makes different assumptions about duration dependence. I use a Weibull specification because it does not assume a functional form of the dependence parameter, but instead allows one to test for the existence of duration dependence. All analyses were run using STATA 5.0.

21

I estimate a hazard model that accounts for the impact of a series of covariates on the

expected duration of a conflict. This model includes time varying covariates as well as time

invariant ones. I control for the characteristics of the conflict, as well as markers identifying

various aspects of any associated intervention.

Data on civil conflicts during the period 1944 through 1999 are used to test the

hypotheses outlined above. These data are an extension of those generated and reported by

Regan (1996; 1998; 2000) and include information on the conflict and any associated

interventions. The extensions involve two dimensions of the earlier data set. First, I extend

the temporal period from 1994 through 1999. Second, I record information about the

intervention along a number of specific categories. The month and year in which an

intervention is undertaken is recorded, as is the type and magnitude of the intervention. For

example, a military intervention into a conflict would be recorded as to whether it involved

one of six categories (troops, naval support, equipment or aid, intelligence or advisors, air

support, or sanctions), and the size of the intervention in each category. Size is recorded in

terms of the number of troops, the dollar value of the aid, the number of weapons, or

whatever the particular metric applicable for the specific category. Additionally, I recorded

interventions in which there was a direct use of force by the intervener. The same type of

information was recorded for economic interventions.

For each particular month that a conflict is ongoing data are recorded on a number

of contextual and structural variables, as well as the initiation of any outside interventions.

There are 150 conflicts of which 49 have not had an outside intervention; 31 of the conflicts

are right censored (or still ongoing as of December 31, 1999). Of the 101 conflicts with

interventions, the number of individual interventions ranges from one to 75. There were a

22

total of 1036 individual interventions in the 100 conflicts.4 For each month that a conflict

was ongoing a data point was recorded, and within each conflict-month data on

characteristics of the conflict and any interventions were identified and recorded. The unit

of analysis is the conflict month. There are a total of 13,048 conflict months in the data.

The longest running conflict spans a total of 616 months; the shortest lasting only one

month.

Outcome Variable

The outcome variable is operationalized as the duration of the conflict in months. Each

conflict month is coded zero if the conflict continues during that month and one if it

terminates during the month. A conflict is considered to have ended when the military

hostilities have ceased for a period of at least six months. Conceptually this frames the

outcome in terms of settlement rather than resolution. Resolving a conflict is a task far

beyond the capabilities of a military or economic intervention, and it would be

unreasonable to attribute the resolution of the issues at stake to an intervention of this form.

A six-month respite from the fighting gives the policy community time to initiate additional

(or alternative) methods of conflict management that may facilitate the resolution of the

conflict. This is also a criterion used to determine the termination of Militarized Interstate

Disputes (Jones, Bremer and Singer, 1996).

4 Data on the conflict in Israel was dropped from the analysis. The Palestinian conflict within the occupied territories of Israel presented unique and insurmountable coding problems. Numerous outside actors intervened in support of either the Palestinian cause or the Israeli’s, but it was nearly impossible to identify each specific intervention and to determine that it satisfied the convention-breaking and authority-targeted criteria. Additionally, support for the Palestinian’s does not always come to or through the PLO, making it particularly difficult to determine what constitutes on intervention. So much of the aid for Israel and the Palestinians reflects ongoing commitments, even though it also seems to be targeted directly to the patterns of authority. In this sense much of what transpires in this conflict meets the authority-targeted criterion but not

23

Predictor Variables

The predictor variables in my model of interventions reflect characteristics of the conflicts

and the intervention. In broad terms they are: 1) whether there was an intervention, 2)

aspects of the timing of the intervention, 3) the number of casualties, 4) the intensity of the

conflict, 5) the type of intervention, and 6) characteristics of the parties to the conflict . To

develop the concept of a strategy for intervening I disaggregate the intervention variables so

as to create indices that reflect the type of third party involvement. I operationalize the

predictor variables as follows:

• The type of intervention is coded on two dimensions. First, whether the intervention involved military or economic instruments. Each individual intervention is recorded and what appear to be simultaneous (or mixed) interventions are given discrete times 1/10 of a month apart. The data in its raw form records the specific makeup of either the military or economic instruments used in the intervention. Second, a dummy variable was created to reflect the use of military force by the intervener.

• Casualties are recorded in aggregate numbers reported for each conflict. This reflects an ex post indicator of the cost of the of the conflict, and I assume that because this is monotonically increasing with time the principles are able to observe this increasing cost.

• Intensity is operationalized as the average number of casualties per month of the conflict.

• The Target of the intervention is operationalized as a dummy variable marking whether the specific intervention supported the government, the opposition, or was neutral. A dichotomous variable reflecting whether the intervention was biased (in support of the government or the opposition) or neutral is used in the analysis. The number of interventions supporting the government was 608, the opposition 378, and there were 50 neutral interventions.

• The timing of an intervention is recorded in terms of the month in a conflict in which it took place.

• Conflict Type was operationalized in terms of the primary identification of the groups in conflict. Three types of conflict were recorded: Ethnic, Religious, and Ideological. Identification of the primary orientation of the groups was based on

the convention-breaking one. See Regan (2000) for a full discussion of the operational definition of an intervention.

24

the Minorities at Risk classification (Gurr, 1995; see Regan, 1996; 2000 for a description). This nominal coding was disaggregated into three dummy variables marking the type of conflict. A variable that reflects ethnic or religious, versus ideological, conflicts was created from the dummy variables.

• A dummy variable records the use of force by the intervener, with the timing of each use of force used to control for the effect of the specific time that force is used in the life of a conflict.

• Data on the ethnic makeup of the society was taken from Ellingsen (2000) and reflects the percentage of the population of the largest ethnic group in the society. Data for Chad, Azerbajian, and Pakistan were based on the size of the largest religious group.

• If the intervention was carried out under the auspices of an international organization a dummy variable records this. There were 107 such interventions.

• Opposing interventions are recorded in terms of interventions that take place in an offsetting sequence. Data are recorded as a marker identifying the existence of an opposing intervention. For example, if an intervention supported the government the next intervention would have to support the opposition for it to be recorded as an opposing intervention. If there are multiple interventions recorded in a given month, and they support opposing actors, each is considered part of the opposing intervention. A neutral intervention is not recorded as opposing a prior intervention by either the government or the opposition.

Results

The baseline hazard function of the probability of a conflict ending at any time ti given that

it is ongoing at t0 can be seen in figure 1. The y-axis represents the probability of a conflict

continuing at any given point, while the x-axis represents the number of months a conflict is

ongoing. We can see that conflicts without third party interventions are quite likely to end

within the first few months of the conflict, with the curve flattening out somewhere around

the first few years of the conflict. Those conflicts with interventions demonstrate a

somewhat different pattern, with a much less pronounced drop off in the probability of a

conflict ending early, with the curves intersecting only after about 15 years of conflict. For

example, the probability of a conflict surviving to the 48th month without an intervention is

about 37%, while with an intervention that probability is about 60%. This suggests that

25

until a conflict becomes rather entrenched the effect of an intervention is to contribute to a

longer running conflict.

[Figure 1 about here]

Table 1 presents the results of the Weibull model of conflict duration with the unit of

analysis being the conflict-month. The results are presented in an accelerated time-failure

metric, and identify the effect of an outside intervention on the expected future duration of a

conflict, when controlling for some of the characteristics that define its scope. The

coefficients associated with an accelerated time failure parameterization can be translated

into expected time until the end of a conflict (Box-Steffensmeier and Jones, 1997). While a

direct interpretation of the results is somewhat problematic, in general a unit change on a

predictor variable has an effect on the expected time until the conflict ends that is consistent

with the sign of the coefficient5. If you exponentiate the coefficient the marginal effect can

be calculated in terms of the percentage longer (or shorter) that we would expect a conflict

to continue. For example, when the conflict is organized around ethno-religious groups in

conflict the hazard that the conflict will end decreases by about 36% less likely than

conflicts organized along ideological lines (e.31). Likewise, The results suggest that

opposing interventions reduce the probability that a conflict will end in the next period by

well beyond 1000%. And contrary to what one might think about the effect of the costs of

the conflict, the total number of fatalities does not influence the expected time until the

conflict ends. An increasing number of fatalities per month, however, does reduce the

5 The actual impact of a change on one of the predictor variables is a function of the timing of all other predictor variables and the time at which the change takes place (see: Teachman and Hayward, 1993)

26

likelihood that a conflict will end sooner (4%). We must evaluate this role of costs

(fatalities and intensity) with caution, however, because both represent ex post indicators of

the cost of continued conflict.

Table 1 Duration Model of Civil Wars: Outcome Time (months) to Terminiation

Model 1 Model 2 ____________________________________________________________________________________

Variable Coeff. S. Error* Stat. Sig. Coeff. S. Error* Stat.Sig Opposing 11.3 1.02 .000 11.7 1.09 .000 Govt X Military --- --- --- -.01 .68 .981 Govt X Economic --- --- --- 11.2 1.45 .000 Opp X Military --- --- --- .53 .50 .298 Neutral X Mil --- --- --- 11.7 1.17 .000 Neutral X Econ --- --- --- 11.3 1.26 .000 Military Interven 11.3 1.17 .000 --- --- --- Econ Interven 21.9 2.05 .000 --- --- --- TimeXForce .018 .016 .278 .018 .015 .258 Use of Force -.53 .79 .504 -.60 .72 .401 Homogeniety -.005 .003 .113 -.005 .003 .116 Fatalities .13 X 10 –5 .50 X 10-6 .007 .13 X 10-5 .5 X 10-6 .008 Intensity .04 .006 .000 .04 .006 .000 Ethno-Religious .31 .12 .011 .30 .12 .013 TimeXGovt -.001 .017 .946 .007 .02 .776 TimeXOpp -.02 .004 .000 -.02 .004 .000 Target (gov/opp) -10.9 1.27 .000 --- --- --- Inter. Organization .04 .42 .921 .34 .43 .432 Intervention .08 .12 .503 .09 .12 .472 Constant 2.52 .31 .000 2.53 .307 .000

Rho = 1.37; σ=.117 Rho= 1.38; σ=.123 *Robust standard errors are used to control for clustering on conflicts Subjects: 150 Subjects: 150 Failures: 119 Failures: 119 Obs: 13048 Obs: 13048 Log Likelihood: -168.97 Log Likelihood: -169.27 Wald Chi2 (14 df): 322.9; p< .000

A conflict with a late intervention on behalf of the opposition is about 2% more

likely to end in the next month than if the intervention was earlier, but the timing of an

27

intervention supporting the government has no effect on expected durations. But regardless

of the character of an intervention – whether it be military or economic – both are

associated with longer conflicts. For example, military support for either the government or

the opposition results in a decreased hazard of the conflict ending in the order of magnitude

of 106 while economic interventions are even less likely to contribute to the end of a

conflict. Interestingly enough, conflicts with biased interventions are considerably more

likely end in the next period than are those with neutral interventions.

Furthermore, the estimate of ‘p’ – the duration dependence parameter – suggests

that the longer a conflict is ongoing the sooner it can be expected to end. That is, civil

conflicts have a positive duration dependence. This shows up as p > 1. This suggests that

civil conflicts wear out the opposing factions. To demonstrate the expected magnitude of

this institutionalizing effect I calculate the decrease in the hazard rate across various points

in time and then compare these hazard rates to each other (see Box-Steffensmeier and

Jones, 1997). The hazard at any specified point can be determined by6:

h(t=d) = expβ0/ρ x ρ x (expβ0/ρ x d) ρ-1

where h(t=d) reflects the time period at which the hazard rate is to be evaluated (d=month),

expβ0/ρ is the constant divided by the duration parameter ρ. Comparing hazard rates at

different periods requires evaluating the hazard at d1 to d2 by the following formula:

h(t=d2) – h(t=d1)/h(t=d1) x 100

Table 2 presents the results of some representative hazard rates from selected time periods.

6 These formulae are derived from Box-Steffensmeier and Jones, 1997.

28

Table 2 Time and the duration of Civil Conflicts

Duration % ∆ in baseline risk 1 month vs. 10 months 1.03% 10 months vs. 100 months 1.66% 100 months vs. 200 months .13% 200 months vs. 300 months .16%

__________________________________________

The interesting interpretation from these results is that risk of a conflict ending between the

10th and the 100th months is 60% greater than in the first ten months, and 150% greater

than in the period between the 100th and 300th months. As the conflict continues it remains

more likely to end in each subsequent month, but at a non-linear rate throughout the life of

the conflict.

Discussion

The main questions addressed by this paper were whether third party interventions tend to

shorten or lengthen the duration of intrastate conflicts, and whether particular strategies for

intervening have different effects on the expected duration of a conflict. The analysis

presented above gives a fairly solid foundation by which to evaluate these questions. First,

if a conflict attracts opposing interventions the likelihood of the conflict ending in the next

month drops precipitously close to zero. Second, it appears that interventions themselves -

- whether military or economic – do not act as effective tools of conflict management. That

is, either type of intervention alone greatly increases the expected duration of a conflict.

Third, the timing of an intervention appears to have little or no independent effect on the

29

duration of a conflict; I will come back to this topic shortly. Fourth, it does make a

difference whether the intervener supports one side in the conflict rather than trying to be

neutral, in the sense that a biased intervention will shorten the expected duration relative to

a neutral intervention. And finally, there are certain characteristics of the conflict that will

affect its expected duration, independent of attempts to manage it through outside

interventions.

I will first review the evidence in light of the articulated hypotheses and then discuss

some of the implications in terms of theoretical development and policy formation.

Hypotheses 1 & 2 posited that the timing and the target of an intervention would

have a direct impact on the likelihood that a conflict will continue into the following month.

Keeping in mind that empirically this was operationalized as a time varying covariate,

meaning that the evidence evaluates the effect of specific interventions at particular points

in time during a conflict, the results do not support the hypotheses. For example, I

hypothesized that an early intervention on behalf of the rebels would lengthen the conflict,

while an early intervention supporting the government would shorten it. In terms of the

effect of an intervention supporting the government, there is no statistically significant

relationship to the duration of a conflict. The evidence relating the timing of an

intervention supporting the opposition is in the anticipated direction and is statistically

significant, but the substantive impact is quite small. So although hypothesis #1 is refuted

and #2 is confirmed, the ability to draw inferences from them is muted by the meager

marginal impact of this strategy for assisting the opposition movement. The evidence is of

course mixed because overwhelmingly any intervention tends to increase the expected

duration of a conflict. A good strategy is not enough. This can be seen in model 2, Table 1

30

where interaction variables reflecting the type and target of an intervention are used to

examine more closely the impact of the strategy of an intervention on the expected duration

of a conflict. Overwhelmingly the conclusion is that regardless of the target or the type, an

intervention tends to decrease the likelihood that a conflict will end in the next month.

Hypotheses three through five postulate that the use of force by the intervener and

the existence of counter interventions will have an impact on the likelihood that a conflict

will end in the next period. That is, opposing interventions are expected to prolong a

conflict while the timely use of force should shorten the conflict. The results, however,

suggest that the use of force has no bearing on whether or not a conflict will end in the next

period, but any conflict that attracts opposing interventions is considerably more likely to

remain ongoing than a conflict that does not have interventions supporting both sides. In

fact this result is so strong and remains robust on all model specifications that one can only

conclude that opposing interventions exacerbate a conflict and cannot be driven by a motive

of conflict management.

Quite surprisingly the evidence does not support hypothesis six. Neutral

interventions are strongly associated with longer conflicts than are biased interventions.

Two points are quite remarkable about this and deserve mention. First, neutral

interventions would appear on their face to be the sine qua non of a policy designed to

manage a conflict, yet the evidence suggests that neutral policies are largely incapable of

contributing to the termination of a conflict. Second, interventions carried out under the

auspices of international organizations have no independent effect on the expected time

until a conflict ends. Conventional wisdom would suggest that 1) there would be a high

degree of overlap between neutrality and multilateral interventions (Diehl, 1995; Regan,

31

2000), and 2) that the perceived impartiality of international organizations would facilitate

the more immediate end to the fighting. There are two potential explanations to account for

the lack of empirical support: international organizations are not always neutral and

neutrality during an ongoing conflict may be insufficient to convince the actors that

stopping the fighting is in their best interest. In short, neutral interventions are tentative

rather than definitive. Furthermore, only about 45% of the interventions by international

organizations were neutral with respect to the parties in the conflict and therefore many of

the instances of multilateral interventions simply reflect a strategy somewhat analogous to

that of a unilateral intervention. Impartiality and multilateral organizational structures may

be more effective at maintaining stability once a conflict has ended, but they do not seem to

be characteristics of an intervention that facilitates the quick end to an ongoing conflict.

The policy implications of these results are fairly stark. If the objective of an

intervention is to shorten the length of a civil conflict, then an outside military or economic

intervention is not a terribly effective strategy to do so. Regardless of how the intervention

is conceived – or empirically operationalized -- there seems to be no mix of strategies that

lead to shorter expected durations.7 Even maintaining a neutral posture or organizing the

intervention under the auspices of a multilateral rubric is not sufficient to form an effective

means of conflict management.

Clearly the notion of what constitutes an effective intervention strategy is an

important element in understanding how civil conflicts are managed. When focusing

exclusively on military and economic forms of outside interventions, conflicts appear to be

not only ineffectively managed, but the interventions themselves appear to prolong the

32

conflict. Further theoretical development must account for the simultaneous impact of

diplomacy as a tool of conflict management, as well as rethinking the assumption –

implicitly adopted here – that the effect of an intervention reflects a proportional hazard rate

over time (Box-Steffensmeier and Zorn, 2001). The results further suggest that

policymakers need to think more critically about the role of military or economic

interventions if their objective is to manage the violent aspects of a conflict.

7The evidence suggests that biased interventions shorten dramatically a conflict relative to a neutral intervention, but this is tempered by the fact that any type of intervention lengthens the time until the conflict ends relative to no intervention.

33

References

Bennett D. Scott and Allan C. Stam III. 1996. “The Duration of Interstate Wars, 1816-

1985.” American Political Science Review 90 (June): 239-257. Bennett, D. S. 1999. “Parametric models, duration dependence, and time-varying data

revisited: comment on J. M. Box-Steffensmeier and B. S. Jones”. American Journal of Political Science, vol. 43 (1) : 256-70.

Bobrow, Davis B. and Mark A. Boyer. 1997. “Maintaining System Stability: Contributions

to Peacekeeping Operations”. Journal of Conflict Resolution, vol. 41, no.6: 723-748. Blossfeld, Hans-Peter, and Götz Rohwer. 1995. Techniques of Event History Modeling.

Mahwah, NJ: Lawrence Erlbaum Associates.

Box-Steffensmeier, Janet M. and Christopher J. W. Zorn. forthcoming. “Duration Models and Proportional Hazards in Political Science”, American Journal of Political Science,

Box-Steffensmeier, Janet M. and Bradford S. Jones. 1997. “Time is of the Essence: Event History Models in Political Science. American Journal of Political Science, vol. 41, no.4:1414-1461.

Collier, Paul, Anke Hoeffler, and Mans Soderbom. 1999. “On the Duration of Civil War”. Typescript. Washington, D.C.: World Bank, Development Economic Research Group.

Diehl, Paul F. 1993. International Peacekeeping. Baltimore: Johns Hopkins University

Press. Ellingsen, Tanja. 2000. “Colorful Community or Ethnic Witches' Brew?: Multiethnicity and

Domestic Conflict During and After the Cold War”. Journal of Conflict Resolution, vol. 44, no. 2:228-249.

Elbadawi, Ibrahim and Nicholas Sambanis. 2000. “External Interventions and the Duration of Civil Wars”. Presented on the workshop on the Economics of Civil Violence, Princeton University, March 18-19, 2000. Elbadawi, Ibrahim A. 1999. “Civil Wars and Poverty: The Role of External Interventions,

Political Rights and Economic Growth”. Presented at the World Bank’s Conference on “Civil Conflicts, Crime and Violence”, Washington, D.C., February 22-23, 1999.

Gurr, Ted Robert. 1993. Minorities at Risk: A Global View of Ethnopolitical Conflicts.

Washington, D. C.: United States Institute for Peace. Haass, Richard N. 1990. Conflicts Unending: The United States and Regional Disputes.

New Haven: Yale.

34

Khanna, Jyoti, Todd Sandler, and Hirofumi Shimizu. 1998. “Sharing the Financial Burden of UN and NATO Peacekeeping, 1976-1996. Journal of Conflict Resolution, vol. 42, no.2: 176-195. Kaufmann, Chaim. 1996. “Possible and Impossible Solution to Ethnic Conflicts”.

International Security. Vol. 20, no.4, pp. 136-175. Lichbach, Mark, 1995. The Rebel’s Dilemma. Ann Arbor, MI: University of Michigan Press. Licklider, Roy. 1995. "The Consequences of Negotiated Settlements in Civil Wars, 1945-

1993". American Political Science Review, vol. 89, no. 3:681-690. Licklider, Roy. 1993."How Civil Wars End: Questions and Methods", in Roy Licklider, ed.

Stopping the Killing: How Civil Wars End. New York: New York University Press. Lindsay, Dylan and Andrew J. Enterline. 1999. “Prolonging the Killing? Third Party

Intervention and the Duration of Intrastate Conflict, 1944-92”. Presented at the Annual Meeting of the Western Political Science Association, Seattle, WA, March 25-27, 1999.

Pillar, Paul R. 1983. Negotiating Peace: War Termination as a Bargaining Process.

Princeton: Princeton University Press. Regan, Patrick M. 1996. "Third Party Interventions into Intra-State Conflicts: Identifying

the Conditions of Successful Intervention Attempts", Journal of Conflict Resolution vol 40. no.1 pp:336-359

Regan, Patrick M. 2000. Civil Wars and Foreign Powers: Outside Interventions and

Intrastate Conflict. Ann Arbor: University of Michigan Press Regan, Patrick M. and Allan C. Stam. 2000. “In the Nick of Time: Conflict Management,

Mediation Timing, and the Duration of Interstate Disputes”. International Studies Quarterly, vol. 44, no. 2: 239-260.

Regan, Patrick M.. 1998 "Choosing to Intervene: Outside Interventions into Internal Conflicts as a

Policy Choice", Journal of Politics, vol. 60. No. 3 Teachman, Jay D. and Mark D. Hayward. 1993. “Interpreting Hazard Rate Models”,

Sociological Methods and Research, vol. 21, no.3:340-71. Walter, Barbara. 2000. “Negotiating Settlements to Civil Wars” unpublished ms. University of California, San Diego.

35

Figure 1Survival Probabilities with and without Interventions

Months of Conflict1 50 100 200 300 400 500 600 700

0.00

0.25

0.50

0.75

1.00

No Interventions

Interventions